Clarification on radar fundamental matrix

[chtanch]

In the paper, in Eq. 13 and 14, the radar intrinsic matrix is an orthographic projection matrix and a conversion to cartesian coordinates/units.

However, in ProjectAllBBox.py, it seems that a perspective transform is applied as well. So there seems to be a contradiction between the code and paper's model + formula. Could you clarify this?

radar_coords = np.dot(P ,world_coords)
radar_coords /= radar_coords[-1]

If I set the world coordinates (x,z) = (0,0) to estimate the translation between world/camera and radar coordinate frames using

radar_coords = np.dot(P, np.array([0,0,1,1]).T)
radar_coords /= radar_coords[-1]

The result is [-0.2125, 0.6 , 1. ] in meters, which does not seem reasonable.

Furthermore, can you clarify the difference between 'p_matrix_original.npy' and 'p_matrix.npy'?

Thank you.